[Phi] Unify dispatch macros to visit (#41653)

* chnage dispatch to visit

* resolve conflict

paddle/phi/api/lib/data_transform.cc

@@ -14,7 +14,6 @@ limitations under the License. */
 
 #include "paddle/phi/api/lib/data_transform.h"
 
-#include "paddle/phi/api/ext/dispatch.h"
 #include "paddle/phi/api/lib/kernel_dispatch.h"
 #include "paddle/phi/api/lib/utils/storage.h"
 #include "paddle/phi/backends/all_context.h"

paddle/phi/kernels/cpu/cast_grad_kernel.cc

@@ -13,7 +13,9 @@
 // limitations under the License.
 
 #include "paddle/phi/kernels/cast_grad_kernel.h"
+
 #include "paddle/phi/core/kernel_registry.h"
+#include "paddle/phi/core/visit_type.h"
 #include "paddle/phi/kernels/cpu/cast_impl.h"
 
 namespace phi {

paddle/phi/kernels/cpu/cast_impl.h

@@ -13,7 +13,7 @@
 // limitations under the License.
 
 #pragma once
-#include "paddle/phi/api/ext/dispatch.h"
+
 #include "paddle/phi/backends/cpu/cpu_context.h"
 
 // See Note [ Why still include the fluid headers? ]

paddle/phi/kernels/cpu/cast_kernel.cc

@@ -16,6 +16,7 @@
 #include "paddle/phi/kernels/cpu/cast_impl.h"
 
 #include "paddle/phi/core/kernel_registry.h"
+#include "paddle/phi/core/visit_type.h"
 
 namespace phi {
 

paddle/phi/kernels/cpu/cross_entropy_grad_kernel.cc

@@ -16,13 +16,11 @@ limitations under the License. */
 
 #include "paddle/phi/backends/cpu/cpu_context.h"
 #include "paddle/phi/core/kernel_registry.h"
+#include "paddle/phi/core/visit_type.h"
 #include "paddle/phi/kernels/copy_kernel.h"
 #include "paddle/phi/kernels/funcs/axis_utils.h"
 #include "paddle/phi/kernels/funcs/eigen/common.h"
 
-// TODO(chenweihang): move dispatch.h into phi/core
-#include "paddle/phi/api/ext/dispatch.h"
-
 namespace phi {
 
 template <typename T, typename LabelT>
@@ -200,7 +198,7 @@ void CrossEntropyWithSoftmaxGradKernel(const Context& dev_ctx,
                                                axis,
                                                logits_grad);
   } else {
-    PD_DISPATCH_INTEGRAL_TYPES(
+    PD_VISIT_INTEGRAL_TYPES(
         dtype, "CrossEntropyWithSoftmaxGradCPUKernel", ([&] {
           CrossEntropyWithSoftmaxGradCPUKernel<T, data_t>(dev_ctx,
                                                           label,

paddle/phi/kernels/cpu/elementwise_kernel.cc

@@ -13,7 +13,6 @@
 // limitations under the License.
 
 #include "paddle/phi/kernels/cpu/elementwise.h"
-#include "paddle/phi/api/ext/dispatch.h"
 #include "paddle/phi/backends/cpu/cpu_context.h"
 #include "paddle/phi/common/bfloat16.h"
 #include "paddle/phi/common/complex.h"

paddle/phi/kernels/cpu/reduce.h

@@ -16,8 +16,8 @@
 
 #include <set>
 
-#include "paddle/phi/api/ext/dispatch.h"
 #include "paddle/phi/backends/cpu/cpu_context.h"
+#include "paddle/phi/core/visit_type.h"
 #include "paddle/phi/kernels/cast_kernel.h"
 
 #include "paddle/phi/api/lib/utils/storage.h"

paddle/phi/kernels/cpu/transpose_kernel.cc

@@ -13,8 +13,9 @@
 // limitations under the License.
 
 #include "paddle/phi/kernels/transpose_kernel.h"
+
 #include <vector>
-#include "paddle/phi/api/ext/dispatch.h"
+
 #include "paddle/phi/backends/cpu/cpu_context.h"
 #include "paddle/phi/common/bfloat16.h"
 #include "paddle/phi/core/kernel_registry.h"

paddle/phi/kernels/funcs/reduce_function.h

@@ -35,7 +35,6 @@ namespace cub = hipcub;
 
 #include "paddle/fluid/platform/device/gpu/gpu_device_function.h"
 #include "paddle/fluid/platform/device/gpu/gpu_launch_config.h"
-#include "paddle/phi/api/ext/dispatch.h"
 #include "paddle/phi/backends/gpu/gpu_context.h"
 #include "paddle/phi/backends/gpu/gpu_info.h"
 #include "paddle/phi/core/dense_tensor.h"

paddle/phi/kernels/gpu/cast_grad_kernel.cu

@@ -12,8 +12,10 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-#include "paddle/phi/core/kernel_registry.h"
 #include "paddle/phi/kernels/cast_grad_kernel.h"
+
+#include "paddle/phi/core/kernel_registry.h"
+#include "paddle/phi/core/visit_type.h"
 #include "paddle/phi/kernels/gpu/cast_impl.h"
 
 namespace phi {

paddle/phi/kernels/gpu/cast_impl.h

@@ -13,7 +13,7 @@
 // limitations under the License.
 
 #pragma once
-#include "paddle/phi/api/ext/dispatch.h"
+
 #include "paddle/phi/backends/gpu/gpu_context.h"
 #include "paddle/phi/kernels/funcs/elementwise_base.h"
 

paddle/phi/kernels/gpu/cast_kernel.cu

@@ -12,8 +12,10 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-#include "paddle/phi/core/kernel_registry.h"
 #include "paddle/phi/kernels/cast_kernel.h"
+
+#include "paddle/phi/core/kernel_registry.h"
+#include "paddle/phi/core/visit_type.h"
 #include "paddle/phi/kernels/gpu/cast_impl.h"
 
 namespace phi {

paddle/phi/kernels/gpu/cross_entropy_grad_kernel.cu

@@ -24,15 +24,13 @@ namespace cub = hipcub;
 
 #include "paddle/phi/common/amp_type_traits.h"
 #include "paddle/phi/core/kernel_registry.h"
+#include "paddle/phi/core/visit_type.h"
 #include "paddle/phi/kernels/copy_kernel.h"
 #include "paddle/phi/kernels/funcs/axis_utils.h"
 #include "paddle/phi/kernels/funcs/for_range.h"
 #include "paddle/phi/kernels/funcs/math_function.h"
 #include "paddle/phi/kernels/gpudnn/softmax_gpudnn.h"
 
-// TODO(chenweihang): move dispatch.h into phi/core
-#include "paddle/phi/api/ext/dispatch.h"
-
 #include "paddle/fluid/operators/math/cross_entropy.h"
 #include "paddle/fluid/operators/math/softmax.h"
 #include "paddle/fluid/platform/device/gpu/gpu_device_function.h"
@@ -267,7 +265,7 @@ void CrossEntropyWithSoftmaxGradKernel(const Context& dev_ctx,
                                                axis,
                                                logits_grad);
   } else {
-    PD_DISPATCH_INTEGRAL_TYPES(
+    PD_VISIT_INTEGRAL_TYPES(
         dtype, "CrossEntropyWithSoftmaxGradGPUKernel", ([&] {
           CrossEntropyWithSoftmaxGradGPUKernel<T, data_t>(dev_ctx,
                                                           label,

paddle/phi/kernels/gpu/cross_entropy_kernel.cu

@@ -24,15 +24,13 @@ namespace cub = hipcub;
 
 #include "paddle/phi/common/amp_type_traits.h"
 #include "paddle/phi/core/kernel_registry.h"
+#include "paddle/phi/core/visit_type.h"
 #include "paddle/phi/kernels/copy_kernel.h"
 #include "paddle/phi/kernels/funcs/axis_utils.h"
 #include "paddle/phi/kernels/funcs/for_range.h"
 #include "paddle/phi/kernels/funcs/math_function.h"
 #include "paddle/phi/kernels/gpudnn/softmax_gpudnn.h"
 
-// TODO(chenweihang): move dispatch.h into phi/core
-#include "paddle/phi/api/ext/dispatch.h"
-
 #include "paddle/fluid/operators/math/cross_entropy.h"
 #include "paddle/fluid/operators/math/softmax.h"
 #include "paddle/fluid/platform/device/gpu/gpu_device_function.h"
@@ -1529,9 +1527,9 @@ void CrossEntropyWithSoftmaxKernel(const Context& dev_ctx,
                                             softmax,
                                             loss);
   } else {
-    PD_DISPATCH_INTEGRAL_TYPES(
-        dtype, "CrossEntropyWithSoftmaxCUDAKernel", ([&] {
-          CrossEntropyWithSoftmaxCUDAKernel<T, data_t>(dev_ctx,
+    PD_VISIT_INTEGRAL_TYPES(dtype, "CrossEntropyWithSoftmaxCUDAKernel", ([&] {
+                              CrossEntropyWithSoftmaxCUDAKernel<T, data_t>(
+                                  dev_ctx,
                                   logits,
                                   label,
                                   soft_label,

paddle/phi/kernels/gpu/reduce.h

@@ -18,6 +18,7 @@
 #if defined(PADDLE_WITH_CUDA) || defined(PADDLE_WITH_HIP) || \
     defined(PADDLE_WITH_XPU_KP)
 
+#include "paddle/phi/core/visit_type.h"
 #include "paddle/phi/kernels/funcs/reduce_function.h"
 
 namespace phi {

paddle/phi/kernels/gpu/reduce_grad.h

@@ -23,7 +23,7 @@
 #include <set>
 #include <vector>
 
-#include "paddle/phi/api/ext/dispatch.h"
+#include "paddle/phi/core/visit_type.h"
 #include "paddle/phi/kernels/funcs/broadcast_function.h"
 
 namespace phi {

paddle/phi/kernels/gpu/transpose_kernel.cu

@@ -14,7 +14,6 @@
 
 #include <vector>
 
-#include "paddle/phi/api/ext/dispatch.h"
 #include "paddle/phi/core/kernel_registry.h"
 #include "paddle/phi/kernels/transpose_kernel.h"
 

paddle/phi/kernels/sparse/cpu/convolution_grad_kernel.cc

@@ -13,13 +13,12 @@ See the License for the specific language governing permissions and
 limitations under the License. */
 
 #include "paddle/phi/kernels/sparse/convolution_grad_kernel.h"
+#include "paddle/phi/core/visit_type.h"
 #include "paddle/phi/kernels/copy_kernel.h"
 #include "paddle/phi/kernels/funcs/blas/blas.h"
 #include "paddle/phi/kernels/funcs/math_function.h"
 #include "paddle/phi/kernels/sparse/cpu/convolution.h"
 
-#include "paddle/phi/api/ext/dispatch.h"
-
 namespace phi {
 namespace sparse {
 
@@ -191,7 +190,7 @@ void Conv3dGradKernel(const Context& dev_ctx,
                       const bool subm,
                       SparseCooTensor* x_grad,
                       DenseTensor* kernel_grad) {
-  PD_DISPATCH_INTEGRAL_TYPES(
+  PD_VISIT_INTEGRAL_TYPES(
       x.non_zero_indices().dtype(), "Conv3dGradCPUKernel", ([&] {
         Conv3dGradCPUKernel<T, data_t>(dev_ctx,
                                        x,

paddle/phi/kernels/sparse/cpu/convolution_kernel.cc

@@ -15,10 +15,9 @@ limitations under the License. */
 #include "paddle/phi/kernels/sparse/cpu/convolution.h"
 #include "paddle/phi/core/kernel_registry.h"
 #include "paddle/phi/core/tensor_meta.h"
+#include "paddle/phi/core/visit_type.h"
 #include "paddle/phi/kernels/funcs/blas/blas.h"
 
-#include "paddle/phi/api/ext/dispatch.h"
-
 namespace phi {
 namespace sparse {
 
@@ -159,7 +158,7 @@ void Conv3dKernel(const Context& dev_ctx,
                   const bool subm,
                   SparseCooTensor* out,
                   DenseTensor* rulebook) {
-  PD_DISPATCH_INTEGRAL_TYPES(
+  PD_VISIT_INTEGRAL_TYPES(
       x.non_zero_indices().dtype(), "Conv3dCPUKernel", ([&] {
         Conv3dCPUKernel<T, data_t>(dev_ctx,
                                    x,

paddle/phi/kernels/sparse/cpu/sparse_mask_kernel.cc

@@ -16,13 +16,12 @@ limitations under the License. */
 #include "paddle/phi/core/ddim.h"
 #include "paddle/phi/core/enforce.h"
 #include "paddle/phi/core/kernel_registry.h"
+#include "paddle/phi/core/visit_type.h"
 #include "paddle/phi/kernels/copy_kernel.h"
 #include "paddle/phi/kernels/empty_kernel.h"
 #include "paddle/phi/kernels/funcs/math_function.h"
 #include "paddle/phi/kernels/funcs/sparse/common_shape.h"
 
-#include "paddle/phi/api/ext/dispatch.h"
-
 namespace phi {
 namespace sparse {
 
@@ -78,7 +77,7 @@ void SparseMaskKernel(const Context& dev_ctx,
                       const DenseTensor& x,
                       const SparseCooTensor& mask,
                       SparseCooTensor* out) {
-  PD_DISPATCH_INTEGRAL_TYPES(
+  PD_VISIT_INTEGRAL_TYPES(
       mask.non_zero_indices().dtype(), "SparseMaskCPUKernel", ([&] {
         SparseMaskCPUKernel<T, data_t>(dev_ctx, x, mask, out);
       }));
@@ -145,7 +144,7 @@ void SparseMaskHelperKernel(const Context& dev_ctx,
                             const SparseCooTensor& x,
                             const DenseTensor& mask_indices,
                             DenseTensor* out) {
-  PD_DISPATCH_INTEGRAL_TYPES(
+  PD_VISIT_INTEGRAL_TYPES(
       x.non_zero_indices().dtype(), "SparseMaskHelperCPUKernel", ([&] {
         SparseMaskHelperCPUKernel<T, data_t>(dev_ctx, x, mask_indices, out);
       }));

paddle/phi/kernels/sparse/cpu/sparse_pool_grad_kernel.cc

@@ -14,13 +14,12 @@ limitations under the License. */
 
 #include "paddle/phi/kernels/sparse/sparse_pool_grad_kernel.h"
 #include "paddle/phi/core/kernel_registry.h"
+#include "paddle/phi/core/visit_type.h"
 #include "paddle/phi/kernels/copy_kernel.h"
 #include "paddle/phi/kernels/empty_kernel.h"
 #include "paddle/phi/kernels/funcs/pooling.h"
 #include "paddle/phi/kernels/funcs/sparse/convolution.h"
 
-#include "paddle/phi/api/ext/dispatch.h"
-
 namespace phi {
 namespace sparse {
 
@@ -82,7 +81,7 @@ void MaxPoolGradKernel(const Context& dev_ctx,
                        const SparseCooTensor& out_grad,
                        const std::vector<int>& kernel_sizes,
                        SparseCooTensor* x_grad) {
-  PD_DISPATCH_INTEGRAL_TYPES(
+  PD_VISIT_INTEGRAL_TYPES(
       x.non_zero_indices().dtype(), "MaxPoolGradCPUKernel", ([&] {
         MaxPoolGradCPUKernel<T, data_t>(
             dev_ctx, x, rulebook, out, out_grad, kernel_sizes, x_grad);

paddle/phi/kernels/sparse/cpu/sparse_pool_kernel.cc

@@ -15,12 +15,11 @@ limitations under the License. */
 #include "paddle/phi/kernels/sparse/sparse_pool_kernel.h"
 #include "paddle/phi/core/kernel_registry.h"
 #include "paddle/phi/core/tensor_meta.h"
+#include "paddle/phi/core/visit_type.h"
 #include "paddle/phi/kernels/funcs/pooling.h"
 #include "paddle/phi/kernels/funcs/sparse/convolution.h"
 #include "paddle/phi/kernels/sparse/cpu/convolution.h"
 
-#include "paddle/phi/api/ext/dispatch.h"
-
 namespace phi {
 namespace sparse {
 
@@ -106,7 +105,7 @@ void MaxPoolKernel(const Context& dev_ctx,
                    const std::vector<int>& strides,
                    SparseCooTensor* out,
                    DenseTensor* rulebook) {
-  PD_DISPATCH_INTEGRAL_TYPES(
+  PD_VISIT_INTEGRAL_TYPES(
       x.non_zero_indices().dtype(), "MaxPoolCPUKernel", ([&] {
         MaxPoolCPUKernel<T, data_t>(dev_ctx,
                                     x,

paddle/phi/kernels/sparse/gpu/convolution.cu.h

@@ -338,7 +338,7 @@ int ProductRuleBook(const Context& dev_ctx,
                     SparseCooTensor* out,
                     std::vector<int>* h_counter,
                     std::vector<int>* h_offsets) {
-  // TODO(zhangkaihuo): use PD_DISPATCH_INTEGRAL_TYPES for secondary dispatch
+  // TODO(zhangkaihuo): use PD_VISIT_INTEGRAL_TYPES for secondary dispatch
   auto indices_dtype = paddle::experimental::CppTypeToDataType<IntT>::Type();
   const int64_t non_zero_num = x.nnz();
   const auto& non_zero_indices = x.non_zero_indices();

paddle/phi/kernels/sparse/gpu/convolution_grad_kernel.cu

@@ -18,14 +18,13 @@ limitations under the License. */
 #include "paddle/phi/backends/gpu/gpu_launch_config.h"
 #include "paddle/phi/core/kernel_registry.h"
 #include "paddle/phi/core/tensor_meta.h"
+#include "paddle/phi/core/visit_type.h"
 #include "paddle/phi/kernels/copy_kernel.h"
 #include "paddle/phi/kernels/funcs/blas/blas.h"
 #include "paddle/phi/kernels/funcs/math_function.h"
 #include "paddle/phi/kernels/sparse/convolution_grad_kernel.h"
 #include "paddle/phi/kernels/sparse/gpu/convolution.cu.h"
 
-#include "paddle/phi/api/ext/dispatch.h"
-
 namespace phi {
 namespace sparse {
 
@@ -249,7 +248,7 @@ void Conv3dGradKernel(const Context& dev_ctx,
                       const bool subm,
                       SparseCooTensor* x_grad,
                       DenseTensor* kernel_grad) {
-  PD_DISPATCH_INTEGRAL_TYPES(
+  PD_VISIT_INTEGRAL_TYPES(
       x.non_zero_indices().dtype(), "Conv3dGradGPUKernel", ([&] {
         Conv3dGradGPUKernel<T, data_t>(dev_ctx,
                                        x,

paddle/phi/kernels/sparse/gpu/convolution_kernel.cu

@@ -15,12 +15,11 @@ limitations under the License. */
 #include "paddle/phi/backends/gpu/gpu_context.h"
 #include "paddle/phi/core/kernel_registry.h"
 #include "paddle/phi/core/tensor_meta.h"
+#include "paddle/phi/core/visit_type.h"
 #include "paddle/phi/kernels/funcs/blas/blas.h"
 #include "paddle/phi/kernels/sparse/convolution_kernel.h"
 #include "paddle/phi/kernels/sparse/gpu/convolution.cu.h"
 
-#include "paddle/phi/api/ext/dispatch.h"
-
 namespace phi {
 namespace sparse {
 
@@ -177,7 +176,7 @@ void Conv3dKernel(const Context& dev_ctx,
                   const bool subm,
                   SparseCooTensor* out,
                   DenseTensor* rulebook) {
-  PD_DISPATCH_INTEGRAL_TYPES(
+  PD_VISIT_INTEGRAL_TYPES(
       x.non_zero_indices().dtype(), "Conv3dGPUKernel", ([&] {
         Conv3dGPUKernel<T, data_t>(dev_ctx,
                                    x,

paddle/phi/kernels/sparse/gpu/sparse_mask_kernel.cu

@@ -19,14 +19,13 @@ limitations under the License. */
 #include "paddle/phi/core/ddim.h"
 #include "paddle/phi/core/enforce.h"
 #include "paddle/phi/core/kernel_registry.h"
+#include "paddle/phi/core/visit_type.h"
 #include "paddle/phi/kernels/copy_kernel.h"
 #include "paddle/phi/kernels/empty_kernel.h"
 #include "paddle/phi/kernels/funcs/math_function.h"
 #include "paddle/phi/kernels/funcs/sparse/common_shape.h"
 #include "paddle/phi/kernels/sparse/sparse_mask_kernel.h"
 
-#include "paddle/phi/api/ext/dispatch.h"
-
 namespace phi {
 namespace sparse {
 
@@ -118,7 +117,7 @@ void SparseMaskKernel(const Context& dev_ctx,
                       const DenseTensor& x,
                       const SparseCooTensor& mask,
                       SparseCooTensor* out) {
-  PD_DISPATCH_INTEGRAL_TYPES(
+  PD_VISIT_INTEGRAL_TYPES(
       mask.non_zero_indices().dtype(), "SparseMaskGPUKernel", ([&] {
         SparseMaskGPUKernel<T, data_t>(dev_ctx, x, mask, out);
       }));
@@ -265,7 +264,7 @@ void SparseMaskHelperKernel(const Context& dev_ctx,
                             const SparseCooTensor& x,
                             const DenseTensor& mask_indices,
                             DenseTensor* out) {
-  PD_DISPATCH_INTEGRAL_TYPES(
+  PD_VISIT_INTEGRAL_TYPES(
       x.non_zero_indices().dtype(), "SparseMaskHelperGPUKernel", ([&] {
         SparseMaskHelperGPUKernel<T, data_t>(dev_ctx, x, mask_indices, out);
       }));

paddle/phi/kernels/sparse/gpu/sparse_pool_grad_kernel.cu

@@ -18,14 +18,13 @@ limitations under the License. */
 #include "paddle/phi/backends/gpu/gpu_info.h"
 #include "paddle/phi/backends/gpu/gpu_launch_config.h"
 #include "paddle/phi/core/kernel_registry.h"
+#include "paddle/phi/core/visit_type.h"
 #include "paddle/phi/kernels/copy_kernel.h"
 #include "paddle/phi/kernels/empty_kernel.h"
 #include "paddle/phi/kernels/funcs/math_function.h"
 #include "paddle/phi/kernels/funcs/pooling.h"
 #include "paddle/phi/kernels/funcs/sparse/convolution.h"
 
-#include "paddle/phi/api/ext/dispatch.h"
-
 namespace phi {
 namespace sparse {
 
@@ -129,7 +128,7 @@ void MaxPoolGradKernel(const Context& dev_ctx,
                        const SparseCooTensor& out_grad,
                        const std::vector<int>& kernel_sizes,
                        SparseCooTensor* x_grad) {
-  PD_DISPATCH_INTEGRAL_TYPES(
+  PD_VISIT_INTEGRAL_TYPES(
       x.non_zero_indices().dtype(), "MaxPoolGradGPUKernel", ([&] {
         MaxPoolGradGPUKernel<T, data_t>(
             dev_ctx, x, rulebook, out, out_grad, kernel_sizes, x_grad);

paddle/phi/kernels/sparse/gpu/sparse_pool_kernel.cu

@@ -16,12 +16,11 @@ limitations under the License. */
 
 #include "paddle/phi/core/kernel_registry.h"
 #include "paddle/phi/core/tensor_meta.h"
+#include "paddle/phi/core/visit_type.h"
 #include "paddle/phi/kernels/funcs/pooling.h"
 #include "paddle/phi/kernels/funcs/sparse/convolution.h"
 #include "paddle/phi/kernels/sparse/gpu/convolution.cu.h"
 
-#include "paddle/phi/api/ext/dispatch.h"
-
 namespace phi {
 namespace sparse {
 
@@ -136,7 +135,7 @@ void MaxPoolKernel(const Context& dev_ctx,
                    const std::vector<int>& strides,
                    SparseCooTensor* out,
                    DenseTensor* rulebook) {
-  PD_DISPATCH_INTEGRAL_TYPES(
+  PD_VISIT_INTEGRAL_TYPES(
       x.non_zero_indices().dtype(), "MaxPoolGPUKernel", ([&] {
         MaxPoolGPUKernel<T, data_t>(dev_ctx,
                                     x,

paddle/phi/kernels/transfer_layout_kernel.cc

@@ -14,9 +14,9 @@ limitations under the License. */
 
 #include "paddle/phi/kernels/transfer_layout_kernel.h"
 
-#include "paddle/phi/api/ext/dispatch.h"
 #include "paddle/phi/backends/all_context.h"
 #include "paddle/phi/core/kernel_registry.h"
+#include "paddle/phi/core/visit_type.h"
 #include "paddle/phi/kernels/funcs/math_function.h"
 
 namespace phi {

paddle/phi/kernels/xpu/full_kernel.cc

@@ -14,13 +14,13 @@
 
 #include "paddle/phi/kernels/full_kernel.h"
 
-#include "paddle/phi/api/ext/dispatch.h"
 #include "paddle/phi/backends/xpu/xpu_context.h"
 #include "paddle/phi/common/bfloat16.h"
 #include "paddle/phi/common/complex.h"
 #include "paddle/phi/common/float16.h"
 #include "paddle/phi/common/scalar.h"
 #include "paddle/phi/core/kernel_registry.h"
+#include "paddle/phi/core/visit_type.h"
 
 // See Note [ Why still include the fluid headers? ]
 #include "paddle/fluid/memory/memcpy.h"